CoQ10 Analysis

Analysis Objective

The ADNI database was analyzed using feature selection techniques to identify associations to Alzheimer Disease. One of the associations that was discovered was between Coenzyme Q10 and AD. Coq10 is a natural supplement often used to help alleviate blood pressure and cognition problems. There was also a study done testing the impact of Coenzyme Q10 on AD called Coenzyme Q10 decreases amyloid pathology and improves behavior in a transgenic mouse model of Alzheimer’s disease.

Link to Paper

Due to the fact that coq10 is reported to impact amyloid pathology, and was uncovered as an important feature from the analysis of ADNI data, a deeper investigation may be warranted. The purpose of this analysis is to Analyse the impact of Coenzyme Q10 on the diagnosis of Alzheimers.

Reading Data

data <- 
  read_csv("/Users/liamf/OneDrive/Documents/ADNI Work R/Data/ADNIMERGE.csv") %>%
  mutate(Diagnosis = recode(DX_bl, EMCI = "MCI", LMCI = "MCI", SMC = "CN")) %>%
  filter(VISCODE == "bl") %>%
  select(RID, Diagnosis, AGE, APOE4, PTEDUCAT, MMSE_bl) %>%
  drop_na()

coq10_drugs <- 
  read_csv("/Users/liamf/OneDrive/Documents/ADNI Work R/Data/coq10_data.csv") %>%
  filter(VISCODE2 %in% c("sc", "bl")) %>%
  select(RID, VISCODE2, CMMED, CMDOSE, CMFREQNC, CMREASON) %>%
  mutate(CMMED = if_else(CMMED %in% c("coq10"), "1", "0")) %>%
  mutate(CMMED = as.integer(CMMED)) %>%
  group_by(RID) %>%
  summarise(coq10 = sum(CMMED)) %>%
  mutate(coq10 = replace(coq10, coq10 == 2, 1))

data <- data %>%
  left_join(coq10_drugs, by = "RID") %>%
  drop_na()

The data tables used in this analysis are:

ADNIMERGE: Baseline diagnoses and basic patient information have been extracted from ADNIMERGE. ADNIMERGE is a table provided by ADNI containing key information from many other merged data tables and serves as a good starting point for data analysis.

  • Includes demographics, neuropsychological testing scores, MRI and PET summaries, CSF measures.

  • Also includes both a baseline diagnosis and diagnosis at each visit.


RECCMEDS: Drug information has been extracted for all study phases from RECCMEDS (Concurrent Medications Log [ADNI1,GO,2,3]).

The drug table was also pre cleaned prior to reading. Coq10 was coded many different ways in the table and these were all recoded to the format of coq10 in Excel.

Frequencies

Testing COQ10’s connection to AD

Multinomial Logistic will be used to test the connection between COQ10 and AD diagnosis.

Null Hypothesis: There is no connection between COQ10 and ADdiagnosis.

Alternative Hypothesis: There is a connection between COQ10 and ADdiagnosis.

Screening for Confoundings

ChiSquared and Anova will be used to test for connections between potential confounding variables and the use of COQ10.

Apoe4

chisq.test(data$coq10, data$APOE4, correct=FALSE)
## 
##  Pearson's Chi-squared test
## 
## data:  data$coq10 and data$APOE4
## X-squared = 1.5099, df = 2, p-value = 0.47

Age

summary(aov(PTEDUCAT ~ coq10, data = data))
##               Df Sum Sq Mean Sq F value Pr(>F)  
## coq10          1     34   34.37   4.518 0.0336 *
## Residuals   2194  16686    7.61                 
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Education

summary(aov(AGE ~ coq10, data = data))
##               Df Sum Sq Mean Sq F value Pr(>F)  
## coq10          1    274   273.8   5.256  0.022 *
## Residuals   2194 114307    52.1                 
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Mini Mental State Exam

summary(aov(MMSE_bl ~ coq10, data = data))
##               Df Sum Sq Mean Sq F value Pr(>F)
## coq10          1     14  14.400   2.017  0.156
## Residuals   2194  15662   7.139

Visualization of Statistically Significant Confounders.

Age and Education are the statistically significant potential confounding variables.

## # A tibble: 2 x 2
##   coq10 avg_age
##   <dbl>   <dbl>
## 1     0    73.4
## 2     1    71.9
## # A tibble: 2 x 2
##   coq10 avg_age
##   <dbl>   <dbl>
## 1     0    16.0
## 2     1    16.5

The effect of the significant confounding variables appears minimal. Significance may be due to sample size. The variables will still be included in the logistic regression model as controls.

Test

data$Diagnosis <- relevel(as.factor(data$Diagnosis), ref = "AD")
test <- multinom(Diagnosis ~ coq10 + AGE + PTEDUCAT, data = as.data.frame(data))
## # weights:  15 (8 variable)
## initial  value 2412.552586 
## iter  10 value 2219.720036
## final  value 2219.716831 
## converged

Diagnosis ~ coq10 + AGE + PTEDUCAT is the model formula. Diagnosis is being modeled by coq10, Age, and Education.

Assumptions

performance::check_collinearity(test)
## # Check for Multicollinearity
## 
## Low Correlation
## 
##      Term  VIF Increased SE Tolerance
##     coq10 2.46         1.57      0.41
##       AGE 2.03         1.43      0.49
##  PTEDUCAT 1.80         1.34      0.56

The assumption of Multicollinearity is not violated in the Multinomial Logistic Regression model. A VIF above 5 is cause for concern and a VIF above 10 is a problem. All the variables of the model have a VIF below 5.

Results

Model Summary

## Call:
## multinom(formula = Diagnosis ~ coq10 + AGE + PTEDUCAT, data = as.data.frame(data))
## 
## Coefficients:
##     (Intercept)     coq10         AGE   PTEDUCAT
## CN    0.3459363 0.4794516 -0.03243010 0.17175486
## MCI   2.1354107 0.1014783 -0.03471276 0.08876448
## 
## Std. Errors:
##     (Intercept)     coq10         AGE   PTEDUCAT
## CN    0.7636302 0.2776533 0.008890669 0.02282739
## MCI   0.7256995 0.2793635 0.008497043 0.02110057
## 
## Residual Deviance: 4439.434 
## AIC: 4455.434

Proportion of those with AD not on Coenzyme Q10:
0.1799611

Proportion of those with AD on Coenzyme Q10:
0.1285714

Hypothesis Test Adjusting for Confounders

Coefficient P Values (Tests hypothesis that coefficient is not 0):

##     (Intercept)      coq10          AGE     PTEDUCAT
## CN  0.650537566 0.08420366 2.646437e-04 5.306866e-14
## MCI 0.003255158 0.71641957 4.402458e-05 2.590891e-05

Conclusion: There is not a statistically significant association between Diagnosis and the use of Coenzyme Q10 when accounting for age and education, a > .05.

Hypothesis Test Not Adjusting for Confounders

## 
##  Pearson's Chi-squared test
## 
## data:  data[["Diagnosis"]] and data[["coq10"]]
## X-squared = 7.348, df = 2, p-value = 0.02537

Conclusion: There is a statistically significant association between Diagnosis and the use of Coenzyme Q10 when confounding variables are ignored, a > .05. More reliability, however, will be given to the test that accounts for confounders.

Thoughts for the future.

Even though the logistic regression model, which accounted for age and education was not significant, the univariate ChiSquared test of Coenzyme Q10 and Diagnosis was. Another note is that the feature selection analysis selected a number of other drugs designed to treat hypertension like Coenzyme Q10. These drugs were Diovan, Lisinopril, and Amlodipine.

Visualization of Other Feature Selected BP Drugs

Conclusion

There seems to be some connection between Alzheimer’s Disease and blood pressure and the treatment of blood pressure may impact AD. This will require further investigation.